Please login to view this media
- Talk
- 21/09/2022
- UK
Automatic Identification of Clinical Landmarks and Calculation of Graf Angles in Newborn Hip Screening: A Pilot Study
Description
In this presentation, Abhinav Singh, an orthopedic registrar and PhD student at the University of Oxford, discusses his research on the automatic identification of clinical landmarks and calculation of Graf angles in newborn hip screening, a pilot study aimed at addressing developmental dysplasia of the hip (DDH). He highlights the clinical significance of early detection and the current shortcomings of the selective screening pathway used in the UK, which has not improved the rates of late diagnosis since 1986.
Abhinav explains the multidisciplinary approach taken to develop AI prediction models intended to enhance diagnostic accuracy and objectivity in ultrasound imaging. He outlines the methodology where 190 2D ultrasound images were analyzed by experts and subsequently processed by a convolutional neural network (CNN) to identify key anatomical landmarks with impressive pixel accuracy.
The results revealed that the AI model was proficient in detecting landmarks with a minimal pixel error and demonstrated an agreement rate with expert diagnoses in 85.7% of cases, achieving a sensitivity of 90% and a specificity of 76%.
Abhinav concludes the presentation by emphasizing the potential of their explainable AI model to improve newborn hip screening processes and the next steps focused on validating and translating these findings into clinical practice.